
import pandas as pd
import numpy as np 
from sklearn.compose import ColumnTransformer
import sklearn.preprocessing as pp
from sklearn.model_selection import train_test_split
import datetime
import matplotlib as plt
import seaborn as sns
from sklearn.linear_model import LinearRegression
from sklearn.tree import DecisionTreeRegressor
from sklearn.ensemble import RandomForestRegressor
from sklearn.ensemble import GradientBoostingRegressor
from sklearn import svm
from sklearn.metrics import mean_squared_error
from sklearn.metrics import r2_score
from sklearn.model_selection import cross_val_score
from sklearn.model_selection import LeaveOneOut
from sklearn.model_selection import GridSearchCV, learning_curve
from sklearn.svm import SVR
import pickle
import matplotlib.pyplot as plt
from sklearn.feature_selection import mutual_info_regression
import plotly.express as px
from sklearn.preprocessing import LabelEncoder
from sklearn.pipeline import Pipeline
from sklearn.feature_selection import mutual_info_regression

import warnings
warnings.filterwarnings("ignore")
warnings.simplefilter(action='ignore', category=FutureWarning)


pd.set_option('display.max_columns', None)


#from google.colab import drive
#drive.mount('/content/drive')


dataPath = "/content/drive/MyDrive/Data Mining/Data"


#master = pd.read_csv(dataPath + "/global.csv")
master = pd.read_csv("global.csv")
master.head()


master.shape

(15763, 70)


## Imbalanced Data Ratio 
master[master["ShareYN"] ==True].shape[0] / master.shape[0]

0.04307555668337246


#from google.colab import drive
#drive.mount('/content/drive')


#s22 = pd.read_csv(dataPath + "/NBA_stat_2022.csv")
s22 = pd.read_csv( "NBA_stat_2022.csv")
s22.head()


master[master["ShareYN"] ==True].shape[0]/41

16.5609756097561


hist_mvp = master.copy()
hist_mvp.info()

<class 'pandas.core.frame.DataFrame'>
RangeIndex: 15763 entries, 0 to 15762
Data columns (total 70 columns):
 #   Column            Non-Null Count  Dtype  
---  ------            --------------  -----  
 0   Player            15763 non-null  object 
 1   Pos               15763 non-null  object 
 2   Age               15763 non-null  int64  
 3   Tm                15763 non-null  object 
 4   Team              15763 non-null  object 
 5   G                 15763 non-null  int64  
 6   GS                15276 non-null  float64
 7   MPG               15763 non-null  float64
 8   FG                15763 non-null  float64
 9   FGA               15763 non-null  float64
 10  FG%               15709 non-null  float64
 11  3P                15763 non-null  float64
 12  3PA               15763 non-null  float64
 13  3P%               13351 non-null  float64
 14  2P                15763 non-null  float64
 15  2PA               15763 non-null  float64
 16  2P%               15674 non-null  float64
 17  eFG%              15709 non-null  float64
 18  FT                15763 non-null  float64
 19  FTA               15763 non-null  float64
 20  FT%               15272 non-null  float64
 21  ORB               15763 non-null  float64
 22  DRB               15763 non-null  float64
 23  TRB               15763 non-null  float64
 24  AST               15763 non-null  float64
 25  STL               15763 non-null  float64
 26  BLK               15763 non-null  float64
 27  TOV               15763 non-null  float64
 28  PF                15763 non-null  float64
 29  PTS               15763 non-null  float64
 30  Season            15763 non-null  object 
 31  Decade            15763 non-null  int64  
 32  Trade             15763 non-null  bool   
 33  MPTot             15763 non-null  float64
 34  PER               15760 non-null  float64
 35  TS%               15716 non-null  float64
 36  3PAr              15709 non-null  float64
 37  FTr               15709 non-null  float64
 38  ORB%              15760 non-null  float64
 39  DRB%              15760 non-null  float64
 40  TRB%              15760 non-null  float64
 41  AST%              15760 non-null  float64
 42  STL%              15760 non-null  float64
 43  BLK%              15760 non-null  float64
 44  TOV%              15725 non-null  float64
 45  USG%              15760 non-null  float64
 46  OWS               15763 non-null  float64
 47  DWS               15763 non-null  float64
 48  WS                15763 non-null  float64
 49  WS/48             15760 non-null  float64
 50  OBPM              15763 non-null  float64
 51  DBPM              15763 non-null  float64
 52  BPM               15763 non-null  float64
 53  VORP              15763 non-null  float64
 54  MVP               15763 non-null  bool   
 55  PastMVP           15763 non-null  bool   
 56  W                 15763 non-null  int64  
 57  L                 15763 non-null  int64  
 58  W/L%              15763 non-null  float64
 59  GB                15763 non-null  float64
 60  PS/G              15763 non-null  float64
 61  PA/G              15763 non-null  float64
 62  SRS               15763 non-null  float64
 63  GBC               15763 non-null  float64
 64  Playoffs          15763 non-null  bool   
 65  Ranking_Conf      15763 non-null  float64
 66  Ranking_League    15763 non-null  float64
 67  MVP_share         15763 non-null  float64
 68  Game_played_prop  15763 non-null  float64
 69  ShareYN           15763 non-null  bool   
dtypes: bool(5), float64(55), int64(5), object(5)
memory usage: 7.9+ MB


def show_feature_vs_YN(feature, df):
    fig = px.scatter(data_frame = df,
               x= feature,
               y='ShareYN',
               color='ShareYN',
               color_discrete_sequence=['blue','gray'], 
               #hover_data={
                   #'Win/Lose': False,
                   #'Player': True, 
                   #'year': True,
                   #'seed': True,
                   #'W/L%': True, 
                   #'W': True}
                )
    fig.update_layout(height=500,
                     title = f"{feature} vs. ShareYN" )
    fig.show()


features = ["FG","WS","OWS","DWS","PTS"]

for feature in features:
    show_feature_vs_YN(feature=feature, df=hist_mvp)


temp = hist_mvp[hist_mvp["ShareYN"]]


def show_feature_vs_share(feature, df):
    fig = px.scatter(data_frame = df,
               x= feature,
               y='MVP_share',
               color_discrete_sequence=['blue','gray'], 
               #hover_data={
                   #'Win/Lose': False,
                   #'Player': True, 
                   #'year': True,
                   #'seed': True,
                   #'W/L%': True, 
                   #'W': True}
                )
    fig.update_layout(height=500,
                     title = f"{feature} vs. MVP Share" )
    fig.show()


features = ["PTS","FG","WS","OWS"]

for feature in features:
    show_feature_vs_share(feature=feature, df=temp)


hist_mvp = hist_mvp.fillna(0)
def calculate_ml_scores(df):
    X = df.copy()
    y = X["MVP_share"]

    X.drop('MVP_share', axis=1, inplace=True)

    # Label encoding for categoricals
    for colname in X.select_dtypes("object"):
        X[colname], _ = X[colname].factorize()

    # All discrete features should now have integer dtypes (double-check this before using MI!)
    discrete_features = X.dtypes == int
    
    mi_scores = mutual_info_regression(X, y)
    mi_scores = pd.Series(mi_scores, name="MI Scores", index=X.columns)
    mi_scores = mi_scores.sort_values(ascending=False)
    return X, y, mi_scores
X, y, mi_scores = calculate_ml_scores(df=hist_mvp)
def plot_mi_scores(scores, figsize):
    scores = scores.sort_values(ascending=True)
    width = np.arange(len(scores))
    ticks = list(scores.index)
    
    fig, ax = plt.subplots(figsize=figsize)
    ax.barh(width, scores)
    
    for index, value in enumerate(scores):
        plt.text(value +0.005 , index, str(round(value,2)))
    
    plt.yticks(width, ticks)    
    plt.title("Mutual Information Scores")
plot_mi_scores(mi_scores, figsize=(20,15))


hist_mvp = hist_mvp.fillna(0)
hist_mvp_before2000 = hist_mvp.loc[(hist_mvp['Decade'] == 1980)|(hist_mvp['Decade']==1990)]
hist_mvp_before2000
X, y, mi_scores = calculate_ml_scores(df=hist_mvp_before2000)

plot_mi_scores(mi_scores, figsize=(20,15))


hist_mvp = hist_mvp.fillna(0)
hist_mvp_after2000 = hist_mvp.loc[(hist_mvp['Decade'] == 2000)|(hist_mvp['Decade']==2010)|(hist_mvp['Decade']==2020)]

X, y, mi_scores = calculate_ml_scores(df=hist_mvp_after2000)

plot_mi_scores(mi_scores, figsize=(20,15))


mvps = hist_mvp[hist_mvp["MVP"]]
mvpsInfo = mvps.describe()
mvpsInfo


mvps = hist_mvp[hist_mvp["MVP"]==False]
mvpsInfo = mvps.describe()
mvpsInfo


train = master.copy()
train.shape

(15763, 70)


#correlations of each features
corr_matrix = train.corr()

#plot heat map
mask = np.zeros_like(corr_matrix)
mask[np.triu_indices_from(mask)] = True
with sns.axes_style("white"):
    f, ax = plt.subplots(figsize=(15, 15))
    ax = sns.heatmap(corr_matrix, mask=mask, vmax=.3, square=True,cmap="RdYlGn")


unusefull_columns = [
    'Team',             #same with TM
    'GS',               #Game started, unuseful info, only partilly filled
    'FG',               #Sum of 2P and 3P, depend on 2P and 3P
    'FGA',              #Sum of 2PA and 3PA
    'FG%',              #Depending of 2P% and 3P%
    '2PA',              #Depending of 2P and 2P%
    '3PA',              #Depending of 3P and 3P%
    'FTA',              #Depending of FT and FT%
    'TRB',              #Sum of ORB and DRB
    'Decade',           #Unusefull
    'W',                #covered by W/L ratio
    'L',                #covered by W/L ratio
    'MPTot',            #covered by MP per Game, and MP total depends on number of game per season
    'GB'                #Highly correlated
]
train = train.drop(columns=unusefull_columns, axis = 1)
train.shape

(15763, 56)


train["Trade"].value_counts()

False    15763
Name: Trade, dtype: int64


# The trade column are all false, which means only players who haven't been traded could be elected to award
# Drop this column 
train = train.drop("Trade", axis = 1)
train.shape

(15763, 55)


#The name of player and team is not useful on training 
train = train.drop("Player", axis = 1)
train = train.drop("Tm", axis =1)
train.shape

(15763, 53)


(train.isna().sum()/len(train)).sort_values(ascending=False).head(20)

3P%      0.153017
FT%      0.031149
2P%      0.005646
eFG%     0.003426
FTr      0.003426
3PAr     0.003426
TS%      0.002982
TOV%     0.002411
TRB%     0.000190
PER      0.000190
USG%     0.000190
BLK%     0.000190
STL%     0.000190
AST%     0.000190
DRB%     0.000190
ORB%     0.000190
WS/48    0.000190
WS       0.000000
DBPM     0.000000
OBPM     0.000000
dtype: float64


train[train["3P%"].isna()].head()
# when 3P% is NaN 3P and 3PA is 0, fill NaN with 0


# All the columns contain NAN are ratio, so fill all of them with 0
train.fillna(0, inplace = True)


(train.isna().sum()/len(train)).sort_values(ascending=False).head(5)

Pos     0.0
AST%    0.0
BLK%    0.0
TOV%    0.0
USG%    0.0
dtype: float64


# convert "Season" from str to int so we can easily indexing on "Season"
# rule: Season 2021 = 2020-2021 / Season 1980 = 1979 - 1980 
train["Season"] = train["Season"].apply(lambda x: int(x[-4:]))


# Only use data after 2000
# train2 =  train[train["Season"]>= 2000]
train2 = train.copy()


# create a list of categorical features
cate = list(train2.columns[(train2.dtypes == object) | (train2.dtypes == bool) ])

# create a list of numeric features ("Seanson" is NOT included since I will use it to split data and I don't want to scale it)
num  = list(train2.columns[~ ((train2.dtypes == object) | (train2.dtypes == bool) | (train2.columns == "Season") ) ])   

scale = pp.StandardScaler() 

def scaler(df):
  scaled = scale.fit_transform(df[num])
  df = df.drop(num,1)
  df[num] = scaled
  return df 

scaled = scaler(train2)


encoder = pp.LabelEncoder()
def encode(data):
    '''function to encode non-null data and replace it in the original data'''
    #retains only non-null values
    nonulls = np.array(data.dropna())
    #reshapes the data for encoding
    impute_reshape = nonulls.reshape(-1,1)
    #encode date
    impute_ordinal = encoder.fit_transform(impute_reshape)
    #Assign back encoded values to non-null values
    data.loc[data.notnull()] = np.squeeze(impute_ordinal)
    return data

# Encoding categorical variables in predictor data set 
for i in cate:
    encode(scaled[i])


# Split dataset into inputs (X) and targets (y)
predict_columns = ["MVP", "MVP_share", "ShareYN"]
X = scaled.drop(columns = predict_columns, axis = 1)
y = scaled[["ShareYN","Season"]]


# Split train & test data
# training set containing data from 2020 - 2017 / test set containing data from 2018 - 2021 

X_train = X[~(X["Season"] >= 2018)]
y_train = y[~(y["Season"] >= 2018)]

X_test = X[(X["Season"] >= 2018)]
y_test = y[(y["Season"] >= 2018)]

# drop "Season" which is no longer being useful 
X_train.drop("Season",1,inplace = True)
y_train.drop("Season",1,inplace = True)
X_test.drop("Season",1,inplace = True)
y_test.drop("Season",1,inplace = True)
X.drop("Season",1,inplace = True)
y.drop("Season",1,inplace = True)

# convert all elements into float for machine learning models 
X_train = X_train.apply(lambda x: x.astype("float64"))
y_train = y_train.apply(lambda x: x.astype("float64"))
X_test = X_test.apply(lambda x: x.astype("float64"))
y_test = y_test.apply(lambda x: x.astype("float64"))
X = X.apply(lambda x: x.astype("float64"))
y = y.apply(lambda x: x.astype("float64"))


# Check dimension
print("X_train:", X_train.shape, "\n","y_train:", y_train.shape, "\n", "X_test:",  X_test.shape, "\n",  "y_test:", y_test.shape)

X_train: (13908, 49) 
 y_train: (13908, 1) 
 X_test: (1855, 49) 
 y_test: (1855, 1)


### Handling Imbalanced Data (SMOTE)
from imblearn.over_sampling import SMOTE
sm = SMOTE(random_state = 42)
X_train, y_train = sm.fit_resample(X_train, y_train)

sm = SMOTE(random_state = 42)
X, y = sm.fit_resample(X, y)


# Check dimension
print("X_train:", X_train.shape, "\n","y_train:", y_train.shape)

X_train: (26558, 49) 
 y_train: (26558, 1)


# Feature Reduction PCA
from sklearn.decomposition import PCA

pca = PCA().fit(X_train)

plt.plot(pca.explained_variance_ratio_.cumsum(), lw=3, color='#087E8B')
plt.title('Cumulative explained variance by number of principal components', size=15)
plt.show()


loadings = pd.DataFrame(
    data=pca.components_.T * np.sqrt(pca.explained_variance_), 
    columns=[f'PC{i}' for i in range(1, len(X_train.columns) + 1)],
    index=X_train.columns
)

pc1_loadings = loadings.sort_values(by='PC1', ascending=False)[['PC1']]
pc1_loadings = pc1_loadings.reset_index()
pc1_loadings.columns = ['Attribute', 'CorrelationWithPC1']

plt.figure(figsize=(15, 10), dpi=80)

plt.bar(x=pc1_loadings['Attribute'], height=pc1_loadings['CorrelationWithPC1'], color='#087E8B')
plt.title('PCA loading scores (first principal component)', size=10)
plt.xticks(rotation='vertical')
plt.show()

#https://medium.com/towards-data-science/3-essential-ways-to-calculate-feature-importance-in-python-2f9149592155 references


pca = PCA(0.95)
pca.fit(X_train)
X_train_pca = pca.transform(X_train)
X_test_pca = pca.transform(X_test)
X_pca    = pca.transform(X)


temp = train.copy()
preproc = ColumnTransformer(transformers=[('num', pp.StandardScaler(), num), ('cat', pp.OneHotEncoder(handle_unknown = 'ignore'), cate)])
preproc.fit(temp)
cl = preproc.transform(temp)
cl.shape

(15763, 60)


pca2 = PCA().fit(cl)
plt.plot(pca2.explained_variance_ratio_.cumsum(), lw=3, color='#087E8B')
plt.title('Cumulative explained variance by number of principal components', size=15)
plt.show()


model = PCA(n_components=15).fit(cl)
cl_pca = model.transform(cl)
cl_pca.shape

(15763, 15)


from sklearn.cluster import KMeans

wcss = []
for i in range(1, 11):
    kmeans = KMeans(n_clusters = i, init = 'k-means++',
                    max_iter = 400, n_init = 10, random_state = 0)
    kmeans.fit(cl_pca)
    wcss.append(kmeans.inertia_)
  
plt.plot(range(1, 11), wcss)
plt.title('Elbow Method')
plt.xlabel('Association')
plt.ylabel('WCSS') #within cluster sum of squares
plt.show()


kmeans = KMeans(n_clusters=3, random_state = 18)
model = kmeans.fit(cl_pca)
clusters = model.predict(cl_pca)
train["Cluster"] = clusters
train["Cluster"].value_counts()

1    5725
2    5438
0    4600
Name: Cluster, dtype: int64


plt.scatter(cl_pca[clusters ==0, 0], cl_pca[clusters ==0, 1], s = 75, c= 'blue')
plt.scatter(cl_pca[clusters ==1, 0], cl_pca[clusters ==1, 1], s = 75, c= 'red')
plt.scatter(cl_pca[clusters ==2, 0], cl_pca[clusters ==2, 1], s = 75, c= 'green')

<matplotlib.collections.PathCollection at 0x7f17deb0b650>


g = train.groupby('Cluster').mean()
c = train['Cluster'].value_counts()

features = ["ORB","DRB","AST","STL","BLK","3P","OWS","DWS","PTS","MVP"]
r = g[features]
r['counts'] =c
r


from sklearn.linear_model import LogisticRegression
from sklearn import metrics
from sklearn.metrics import accuracy_score, confusion_matrix, precision_score, recall_score, f1_score,roc_auc_score, roc_curve, classification_report
from sklearn.linear_model import LogisticRegressionCV
from sklearn.model_selection import GridSearchCV, KFold, cross_val_score


# Use grid search to find the optimal combination of hyperparameters in Logistic Regression
grid={"C":np.logspace(-3,3,7), "penalty":["l1","l2"]} # l1 lasso l2 ridge
logreg=LogisticRegression()
logreg_cv=GridSearchCV(logreg,grid,cv=10)
logreg_cv.fit(X,y)

print("tuned hpyerparameters :(best parameters) ",logreg_cv.best_params_)
print("accuracy :",logreg_cv.best_score_)

tuned hpyerparameters :(best parameters)  {'C': 1000.0, 'penalty': 'l2'}
accuracy : 0.9539917479112615


# Logistic Regression 
lr_best = LogisticRegression(penalty = 'l2', C = 1000) #penalty = 'l2', C = 1000

# fit the model with data
lr_best.fit(X_train_pca,y_train)
y_pred=lr_best.predict(X_test_pca)

# Evaluation 
print("Accuracy:",round(metrics.accuracy_score(y_test, y_pred)*100,2),"%")
print("Precision:",round(metrics.precision_score(y_test, y_pred)*100,2),"%")
print("Recall:",round(metrics.recall_score(y_test, y_pred)*100,2),"%")
print("F1 Score:", round(f1_score(y_test,y_pred)*100,2),"%")

Accuracy: 95.42 %
Precision: 37.04 %
Recall: 100.0 %
F1 Score: 54.05 %


y_pred_proba = lr_best.predict_proba(X_test_pca)[::,1]

fpr, tpr, _ = metrics.roc_curve(y_test, y_pred_proba)
auc = metrics.roc_auc_score(y_test, y_pred_proba)
plt.plot(fpr,tpr,label="data 1, auc="+str(auc))
plt.legend(loc=4)
plt.show()


# import the metrics class
from sklearn import metrics
cnf_matrix = metrics.confusion_matrix(y_test, y_pred)
cnf_matrix

# import required modules
import numpy as np
import matplotlib.pyplot as plt
import seaborn as sns
%matplotlib inline

class_names=[0,1] # name  of classes
fig, ax = plt.subplots()
tick_marks = np.arange(len(class_names))
plt.xticks(tick_marks, class_names)
plt.yticks(tick_marks, class_names)
# create heatmap

sns.heatmap(pd.DataFrame(cnf_matrix), annot=True, cmap="YlGnBu" ,fmt='g')
ax.xaxis.set_label_position("top")
plt.tight_layout()
plt.title('Confusion matrix', y=1.1)
plt.ylabel('Actual label')
plt.xlabel('Predicted label')
plt.show()


# 10 folds cross validation 

kfold = KFold(n_splits=10, random_state=2022, shuffle=True)

lr_cv = LogisticRegression(penalty = 'l2', C = 1000)

results = cross_val_score(lr_cv, X_pca, y, cv=kfold)

# Output the accuracy. Calculate the mean and std across all folds. 
print("Accuracy: %.3f%% (%.3f%%)" % (results.mean()*100.0, results.std()*100.0))

Accuracy: 94.879% (0.306%)


from sklearn import svm


clf_svm = svm.SVC()
clf_svm.fit(X_train_pca,y_train)
y_pred_svm = clf_svm.predict(X_test_pca)

print("Accuracy:",round(metrics.accuracy_score(y_test, y_pred_svm)*100,2),"%")
print("Precision:",round(metrics.precision_score(y_test, y_pred_svm)*100,2),"%")
print("Recall:",round(metrics.recall_score(y_test, y_pred_svm)*100,2),"%")
print("F1 Score:", round(f1_score(y_test,y_pred_svm)*100,2),"%")

Accuracy: 97.25 %
Precision: 49.47 %
Recall: 94.0 %
F1 Score: 64.83 %


from sklearn.model_selection import GridSearchCV
from sklearn.model_selection import RandomizedSearchCV
from sklearn.ensemble import RandomForestClassifier


clf_rf = RandomForestClassifier(n_estimators=100)
clf_rf.fit(X_train_pca, y_train)
y_pred_random_forest = clf_rf.predict(X_test_pca)
acc_random_forest = round(clf_rf.score(X_test_pca, y_test) * 100, 2)

print("Accuracy:",round(metrics.accuracy_score(y_test, y_pred_random_forest)*100,2),"%")
print("Precision:",round(metrics.precision_score(y_test, y_pred_random_forest)*100,2),"%")
print("Recall:",round(metrics.recall_score(y_test, y_pred_random_forest)*100,2),"%")
print("F1 Score:", round(f1_score(y_test,y_pred_random_forest)*100,2),"%")

Accuracy: 98.33 %
Precision: 63.77 %
Recall: 88.0 %
F1 Score: 73.95 %


test = s22.copy()


player = s22[["Player"]]


unusefull_columns = [
    'Team',             #same with TM
    'GS',               #Game started, unuseful info, only partilly filled
    'FG',               #Sum of 2P and 3P, depend on 2P and 3P
    'FGA',              #Sum of 2PA and 3PA
    'FG%',              #Depending of 2P% and 3P%
    '2PA',              #Depending of 2P and 2P%
    '3PA',              #Depending of 3P and 3P%
    'FTA',              #Depending of FT and FT%
    'TRB',              #Sum of ORB and DRB
    'Decade',           #Unuseful
    'W',                #covered by W/L ratio
    'L',                #covered by W/L ratio
    'MPTot',            #covered by MP per Game, and MP total depends on number of game per season
    'GB',               #Highly correlated
    'Trade',            #The trade column are all false, which means only players who haven't been traded could be elected to award
    'Player',           #The name of player is not useful 
    'Tm',               #The name of team is not useful 
    'Season'            #Unuseful
]
test = test.drop(columns=unusefull_columns, axis = 1)


# Filling missing values 
test.fillna(0, inplace = True)
(test.isna().sum()/len(test)).sort_values(ascending=False).head(5)

Pos     0.0
DRB%    0.0
AST%    0.0
STL%    0.0
BLK%    0.0
dtype: float64


# create a list of categorical features
cate = list(test.columns[(test.dtypes == object) | (test.dtypes == bool) ])

# create a list of numeric features ("Seanson" is NOT included since I will use it to split data and I don't want to scale it)
num  = list(test.columns[~ ((test.dtypes == object) | (test.dtypes == bool) ) ])   

# scale numeric variables 
scaled2 = scaler(test)

# Encoding categorical variables 
for i in cate:
    encode(scaled2[i])

# PCA dimension reduction
reduced = pca.transform(scaled2)


# predict mvp candidates in the recent season 2021-2022 
y_pred = lr_best.predict(reduced)

# get the player names 
results_22 = player.assign(Prediction = y_pred)

# get confidence score for prediction e.g. the higher the confidence score is, the more likely this player will be predicted as a MVP candidates
results_22['confidence_score'] = lr_best.decision_function(reduced)

# get all the predicted MVP candidates 
candidates = list(results_22[results_22["Prediction"] == 1]["Player"].values)

# display the predicted MVP candidates in descending confidence score 
results_22.sort_values('confidence_score', ascending = False, ignore_index = True).head(15)


results_22[results_22["Prediction"] == 1.0].sort_values('confidence_score', ascending = False, ignore_index = True).shape

(58, 3)


print("number of candidates in average:", master[master["ShareYN"] == True].shape[0] / 41)

number of candidates in average: 16.5609756097561


results_22[results_22["Prediction"] == 1.0].sort_values('confidence_score', ascending = False, ignore_index = True)["Player"]

0       Giannis Antetokounmpo
1                Nikola Jokić
2                 Joel Embiid
3                 Luka Dončić
4                   Ja Morant
5                LeBron James
6               Stephen Curry
7                  Chris Paul
8               DeMar DeRozan
9                Jimmy Butler
10                 Trae Young
11         Karl-Anthony Towns
12               Kevin Durant
13               Jayson Tatum
14               Devin Booker
15           Donovan Mitchell
16                Bam Adebayo
17                Rudy Gobert
18               Jrue Holiday
19              Pascal Siakam
20            Dejounte Murray
21            Khris Middleton
22             Darius Garland
23             Nikola Vučević
24              Anthony Davis
25                Zach LaVine
26              Fred VanVleet
27                 Kyle Lowry
28                Paul George
29              Julius Randle
30              Deandre Ayton
31               Jaylen Brown
32                LaMelo Ball
33              Jarrett Allen
34          Russell Westbrook
35             Draymond Green
36               Bradley Beal
37            Robert Williams
38                Tyler Herro
39          Jaren Jackson Jr.
40               Kyrie Irving
41          Jonas Valančiūnas
42              Tobias Harris
43           D'Angelo Russell
44                Evan Mobley
45             Brandon Ingram
46               Steven Adams
47                Mike Conley
48                 Al Horford
49    Shai Gilgeous-Alexander
50             Andrew Wiggins
51              Miles Bridges
52            Anthony Edwards
53               JaVale McGee
54              Jalen Brunson
55              Mikal Bridges
56               Bobby Portis
57             Scottie Barnes
Name: Player, dtype: object


data_copy =  pd.read_csv(dataPath + "/global.csv")
players = data_copy["Player"]
scaled["Player"] = players


from tqdm import tqdm
from sklearn.metrics import mean_absolute_error


df_candidate = scaled[scaled.ShareYN == 1].drop(columns=['MVP', 'ShareYN']).reset_index()
df_candidate.drop("index",1,inplace= True)


# Split dataset into inputs (X) and targets (y)
X = df_candidate.drop(["MVP_share","Player"], axis = 1)
y = df_candidate[["MVP_share","Season"]]


# Split train & test data
# training set containing data from 2020 - 2017 / test set containing data from 2018 - 2021 

X_train = X[~(X["Season"] >= 2018)]
y_train = y[~(y["Season"] >= 2018)]
X_test = X[(X["Season"] >= 2018)]
y_test = y[(y["Season"] >= 2018)]

# drop "Season" which is no longer being useful 
X_train.drop("Season",1,inplace = True)
y_train.drop("Season",1,inplace = True)
X_test.drop("Season",1,inplace = True)
y_test.drop("Season",1,inplace = True)
X.drop("Season",1,inplace = True)
y.drop("Season",1,inplace = True)

# convert all elements into float for machine learning models 
X_train = X_train.apply(lambda x: x.astype("float64"))
y_train = y_train.apply(lambda x: x.astype("float64"))
X_test = X_test.apply(lambda x: x.astype("float64"))
y_test = y_test.apply(lambda x: x.astype("float64"))
X = X.apply(lambda x: x.astype("float64"))
y = y.apply(lambda x: x.astype("float64"))


pca = PCA(0.95)
pca.fit(X_train)
X_train_pca = pca.transform(X_train)
X_test_pca  = pca.transform(X_test)
X_pca       = pca.transform(X)


# The evaluation metrics would be Root Mean Squared Error and R-squared 
def evaluate(real,pred):
    mse = metrics.mean_squared_error(real,pred)
    rmse = np.sqrt(mse)
    print("Root Mean Squared Error:", rmse, '\n', "R_squared:", metrics.r2_score(real,pred))


from sklearn.linear_model import LinearRegression

lr = LinearRegression()
lr.fit(X_train_pca, y_train)
y_pred_lr = lr.predict(X_test_pca)


evaluate(y_test,y_pred_lr)

Root Mean Squared Error: 3.2111788286763012 
 R_squared: 0.5473208659393916


clf = RandomForestRegressor(random_state=2022)
clf.fit(X_train_pca, y_train)
y_pred_random_forest = clf.predict(X_test_pca)
evaluate(y_test,y_pred_random_forest)

Root Mean Squared Error: 2.6133435663146285 
 R_squared: 0.7001842042921298


from sklearn.model_selection import RandomizedSearchCV
# Number of trees in random forest
n_estimators = [int(x) for x in np.linspace(start = 200, stop = 2000, num = 10)]
# Number of features to consider at every split
max_features = ['auto', 'sqrt']
# Maximum number of levels in tree
max_depth = [int(x) for x in np.linspace(10, 110, num = 11)]
max_depth.append(None)
# Minimum number of samples required to split a node
min_samples_split = [2, 5, 10]
# Minimum number of samples required at each leaf node
min_samples_leaf = [1, 2, 4]
# Method of selecting samples for training each tree
bootstrap = [True, False]
# Create the random grid
random_grid = {'n_estimators': n_estimators,
               'max_features': max_features,
               'max_depth': max_depth,
               'min_samples_split': min_samples_split,
               'min_samples_leaf': min_samples_leaf,
               'bootstrap': bootstrap}


# Use the random grid to search for best hyperparameters
# First create the base model to tune
rf = RandomForestRegressor()
# Random search of parameters, using 3 fold cross validation, 
# search across 100 different combinations, and use all available cores
rf_random = RandomizedSearchCV(estimator = rf, param_distributions = random_grid, n_iter = 100, cv = 3, verbose=2, random_state=42, n_jobs = -1)
# Fit the random search model
rf_random.fit(X_train_pca,y_train)

Fitting 3 folds for each of 100 candidates, totalling 300 fits

RandomizedSearchCV(cv=3, estimator=RandomForestRegressor(), n_iter=100,
                   n_jobs=-1,
                   param_distributions={'bootstrap': [True, False],
                                        'max_depth': [10, 20, 30, 40, 50, 60,
                                                      70, 80, 90, 100, 110,
                                                      None],
                                        'max_features': ['auto', 'sqrt'],
                                        'min_samples_leaf': [1, 2, 4],
                                        'min_samples_split': [2, 5, 10],
                                        'n_estimators': [200, 400, 600, 800,
                                                         1000, 1200, 1400, 1600,
                                                         1800, 2000]},
                   random_state=42, verbose=2)


rf_random.best_params_

{'bootstrap': True,
 'max_depth': 10,
 'max_features': 'auto',
 'min_samples_leaf': 4,
 'min_samples_split': 5,
 'n_estimators': 200}


# Best 
rf_best = RandomForestRegressor(bootstrap = True, max_depth = 10, max_features = 'auto',
                            min_samples_leaf = 4, min_samples_split = 5, n_estimators = 200, random_state=2022)
rf_best.fit(X_train_pca, y_train)
y_pred_random_forest = rf_best.predict(X_test_pca)
evaluate(y_test,y_pred_random_forest)

Root Mean Squared Error: 2.5463424548136784 
 R_squared: 0.7153605347570327


from sklearn.neighbors import KNeighborsRegressor

reg_kNN = KNeighborsRegressor()

param_kNN = {
                'n_neighbors' : np.array([5, 10, 15, 20, 25]),
            }

reg_kNN_CV = GridSearchCV(reg_kNN, param_kNN, cv = 5 )

reg_kNN_CV.fit(X_train_pca, y_train)

reg_kNN_CV.best_estimator_

KNeighborsRegressor(n_neighbors=10)


knn_best = reg_kNN_CV.best_estimator_
knn_best.fit(X_train_pca, y_train)

y_pred_knn = knn_best.predict(X_test_pca)
evaluate(y_test,y_pred_knn)

Root Mean Squared Error: 3.3504665336858457 
 R_squared: 0.5071984542493211


from xgboost import XGBRegressor
gb = XGBRegressor(random_state =2022)
gb.fit(X_train_pca, y_train)
y_pred_gb = gb.predict(X_test_pca)
evaluate(y_test,y_pred_gb)

[21:45:47] WARNING: /workspace/src/objective/regression_obj.cu:152: reg:linear is now deprecated in favor of reg:squarederror.
Root Mean Squared Error: 2.547536768280885 
 R_squared: 0.7150934626911837


xgb1 = XGBRegressor()
parameters = {'nthread':[4], #when use hyperthread, xgboost may become slower
              'objective':['reg:linear'],
              'learning_rate': [.03, 0.05, .07], #so called `eta` value
              'max_depth': [5, 6, 7],
              'min_child_weight': [4],
              'silent': [1],
              'subsample': [0.7],
              'colsample_bytree': [0.7],
              'n_estimators': [500]}

xgb_grid = GridSearchCV(xgb1,
                        parameters,
                        cv = 2,
                        n_jobs = 5,
                        verbose=True)

xgb_grid.fit(X_train_pca, y_train)

print(xgb_grid.best_score_)
print(xgb_grid.best_params_)

Fitting 2 folds for each of 9 candidates, totalling 18 fits
0.5643259845524768
{'colsample_bytree': 0.7, 'learning_rate': 0.03, 'max_depth': 5, 'min_child_weight': 4, 'n_estimators': 500, 'nthread': 4, 'objective': 'reg:linear', 'silent': 1, 'subsample': 0.7}


xgb_best = xgb_grid.best_estimator_
xgb_best.fit(X_train_pca, y_train)
y_pred_gb = xgb_best.predict(X_test_pca)
evaluate(y_test,y_pred_gb)

Root Mean Squared Error: 2.4635326776395634 
 R_squared: 0.7335730534409286


to_drop = ["Player"]

def train_test_split_by_year(year, df, scaling=False):
    #test year = selected year, train year = other years outside of selected year 
    train_df = df[df['Season'] != year]
    test_df = df[df['Season'] == year]
    
    train_df2 = train_df.copy()
    test_df2 = test_df.copy()
    
    train_df2.drop(to_drop, axis=1, inplace=True)
    test_df2.drop(to_drop, axis=1, inplace=True)
    
    if scaling == True:
        sc_X = StandardScaler()
        sc_y = StandardScaler()
        train_df2 = sc_X.fit_transform(train_df2)
        test_df2 = sc_y.fit_transform(test_df2)
    
    X_train = train_df2.copy()
    y_train = X_train["MVP_share"]
    
    X_test = test_df2.copy()
    y_test = X_test["MVP_share"]

    X_train.drop('MVP_share', axis=1, inplace=True)
    cols = X_train.columns
    X_test.drop('MVP_share', axis=1, inplace=True)

    X_train = np.array(X_train)
    y_train = np.array(y_train)
    X_test = np.array(X_test)
    y_test = np.array(y_test)
    
    return X_train, y_train, X_test, y_test, cols


  def run_model(regressor, X_train, y_train, X_test, y_test, df, year):

    #pca = PCA(0.99)
    #pca.fit(X_train)
    #X_train = pca.transform(X_train)
    #X_test  = pca.transform(X_test)

    model = regressor
    model.fit(X_train, y_train) 
    predictions = model.predict(X_test)
    mae = mean_absolute_error(predictions, y_test)
    r2 = r2_score(y_test, predictions)
    
    mvp_race = df[df['Season'] == year]
    mvp_race['predicted_share'] = predictions
    mvp_race = mvp_race.sort_values(["MVP_share", "predicted_share"], ascending = (False, False))
                                    
    actual_winner = mvp_race[mvp_race['MVP_share'] == mvp_race['MVP_share'].max()]['Player']
    predicted_winner = mvp_race[mvp_race['predicted_share'] == mvp_race['predicted_share'].max()]['Player']
    return model, mae, r2, predicted_winner.iloc[0], actual_winner.iloc[0], mvp_race


years = [year for year in range(1980, 2022)]

def run_model_average(df, regressor, print_metrics=False):
    mae_lst = []
    r2_lst = []
    predicted_lst = []
    actual_lst = []
    label_lst =[]
    model_lst = []
    for year in tqdm(years):
        X_train, y_train, X_test, y_test, cols = train_test_split_by_year(year=year, df=df)
        model, mae, r2, predicted_winner, actual_winner, mvp_race = run_model(regressor,
                                                              X_train,
                                                              y_train,
                                                              X_test,
                                                              y_test,
                                                              df=df,
                                                              year=year
                                                            )
        if predicted_winner == actual_winner:
            label = 'correct'
        else:
            label = 'incorrect'
        mae_lst.append(mae)
        r2_lst.append(r2)
        predicted_lst.append(predicted_winner)
        actual_lst.append(actual_winner)
        label_lst.append(label)
        model_lst.append(model)
    d = {
    'year': years,
    'MAE': mae_lst,
    'R squared': r2_lst,
    'Predicted MVP': predicted_lst,
    'Actual MVP': actual_lst,
    'Label': label_lst
    }

    summary = pd.DataFrame(d)
    correct_count = summary['Label'].value_counts().iloc[0]
    incorrect_count = summary['Label'].value_counts().iloc[1]
    accuracy = correct_count / (correct_count + incorrect_count)
    avg_mae = summary['MAE'].mean()
    avg_r2  = summary['R squared'].mean()
    
    if print_metrics == True:
        print(f"Average MAE: {avg_mae}")
        print(f"Average R squared: {avg_r2}")
        print(f"Prediction accuracy: {accuracy}")
    return avg_mae, avg_r2, accuracy, summary, model_lst, cols


lr_avg_mae, lr_avg_r2, lr_accuracy, lr_summary, lr_models, cols = run_model_average(df=df_candidate,
                  regressor = lr,
                  #scaling=True,
                 print_metrics=True)

100%|██████████| 42/42 [00:01<00:00, 26.83it/s]

Average MAE: 2.193944764833127
Average R squared: 0.5455488437921744
Prediction accuracy: 0.6666666666666666


rf_avg_mae, rf_avg_r2, rf_accuracy, rf_summary, rf_models, rf_cols = run_model_average(df=df_candidate,
                  regressor=RandomForestRegressor(bootstrap = True, max_depth = 100, max_features = 'auto', min_samples_leaf = 4,
                                                  min_samples_split = 5, n_estimators =  1000, random_state = 2022),
                print_metrics=True)

100%|██████████| 42/42 [08:23<00:00, 11.98s/it]

Average MAE: 1.6715026928882832
Average R squared: 0.622679428472982
Prediction accuracy: 0.6904761904761905


rf_summary[["year","Predicted MVP","Actual MVP","Label"]].set_index("year")


rf_avg_mae, rf_avg_r2, rf_accuracy, rf_summary, rf_models, rf_cols = run_model_average(df=df_candidate,
                  regressor=rf_best,
                print_metrics=True)

100%|██████████| 42/42 [01:33<00:00,  2.22s/it]

Average MAE: 1.6836810695021451
Average R squared: 0.6194808127150092
Prediction accuracy: 0.6904761904761905


from xgboost import XGBRegressor
xgb_avg_mae, xgb_avg_r2, xgb_accuracy, xgb_summary, xgb_models, xgb_cols = run_model_average(df=df_candidate,
                  regressor = xgb_best , print_metrics=True)

100%|██████████| 42/42 [00:55<00:00,  1.33s/it]

Average MAE: 1.5939746620745443
Average R squared: 0.6469721864974994
Prediction accuracy: 0.6666666666666666


## KNN
rf_avg_mae, rf_avg_r2, rf_accuracy, rf_summary, rf_models, rf_cols = run_model_average(df=df_candidate,
                  regressor=knn_best,
                print_metrics=True)

100%|██████████| 42/42 [00:00<00:00, 47.79it/s]

Average MAE: 2.1088438008465706
Average R squared: 0.4499502094937031
Prediction accuracy: 0.5


test = s22.copy()


# Filling missing values 
test.fillna(0, inplace = True)
(test.isna().sum()/len(test)).sort_values(ascending=False).head(5)

# Drop unuseful columns 
test = test.drop(columns=unusefull_columns, axis = 1)


# create a list of categorical features
cate = list(test.columns[(test.dtypes == object) | (test.dtypes == bool) ])

# create a list of numeric features ("Seanson" is NOT included since I will use it to split data and I don't want to scale it)
num  = list(test.columns[~ ((test.dtypes == object) | (test.dtypes == bool) ) ])   

# scale numeric variables 
scaled2 = scaler(test)

# Encoding categorical variables 
for i in cate:
    encode(scaled2[i])

# PCA dimension reduction
reduced = pca.transform(scaled2)


reduced.shape

(510, 14)


# predict mvp candidates in the recent season 2021-2022 
y_pred = rf_best.predict(reduced)

# get the player names 
results_22 = player.assign(Prediction = y_pred)

# get confidence score for prediction e.g. the higher the confidence score is, the more likely this player will be predicted as a MVP candidates
#results_22['confidence_score'] = lr_best.decision_function(reduced)

# get all the predicted MVP candidates 
candidates = list(results_22[results_22["Prediction"] == 1]["Player"].values)

# display the predicted MVP candidates in descending confidence score 
#results_22.sort_values('confidence_score', ascending = False, ignore_index = True).head(15)


results_22.sort_values("Prediction",ascending = False).head()


y_pred = knn_best.predict(reduced)
results_22 = player.assign(Prediction = y_pred)
candidates = list(results_22[results_22["Prediction"] == 1]["Player"].values)


results_22.sort_values("Prediction",ascending = False).head()


y_pred = xgb_best.predict(reduced)
results_22 = player.assign(Prediction = y_pred)
candidates = list(results_22[results_22["Prediction"] == 1]["Player"].values)


results_22.sort_values("Prediction",ascending = False).head()


y_pred = lr.predict(reduced)
results_22 = player.assign(Prediction = y_pred)
candidates = list(results_22[results_22["Prediction"] == 1]["Player"].values)


results_22.sort_values("Prediction",ascending = False).head()

	Player	Pos	Age	Tm	Team	G	GS	MPG	FG	FGA	FG%	3P	3PA	3P%	2P	2PA	2P%	eFG%	FT	FTA	FT%	ORB	DRB	TRB	AST	STL	BLK	TOV	PF	PTS	Season	Decade	Trade	PastMVP	MPTot	PER	TS%	3PAr	FTr	ORB%	DRB%	TRB%	AST%	STL%	BLK%	TOV%	USG%	OWS	DWS	WS	WS/48	OBPM	DBPM	BPM	VORP	W	L	W/L%	GB	PS/G	PA/G	SRS	GBC	Ranking_Conf	Playoffs	Ranking_League	Game_played_prop
0	Precious Achiuwa	C	22	TOR	Toronto Raptors (7)	59	25	23.1	3.4	7.9	0.429	0.6	1.6	0.362	2.8	6.3	0.447	0.466	1.1	1.8	0.590	2.1	4.7	6.8	1.1	0.5	0.6	1.1	2.1	8.4	2021-2022	2020	False	False	1365.0	12.5	0.484	0.201	0.224	9.3	23.4	16.0	6.9	1.2	2.4	11.2	18.1	0.2	1.6	1.8	0.064	-2.4	-0.7	-3.1	-0.4	38.0	30.0	0.559	16.0	108.9	107.3	1.65	7.0	7	True	14	0.867647
1	OG Anunoby	SF	24	TOR	Toronto Raptors (7)	42	42	36.7	6.6	15.1	0.436	2.4	6.8	0.351	4.2	8.3	0.506	0.515	1.9	2.5	0.748	1.6	3.9	5.5	2.6	1.5	0.6	1.8	3.0	17.5	2021-2022	2020	False	False	1541.0	14.5	0.538	0.449	0.169	4.4	12.3	8.1	10.9	2.1	1.5	9.8	20.9	1.4	1.6	3.0	0.093	0.1	0.0	0.1	0.8	38.0	30.0	0.559	16.0	108.9	107.3	1.65	7.0	7	True	14	0.617647
2	Dalano Banton	SF	22	TOR	Toronto Raptors (7)	58	1	11.5	1.4	3.4	0.403	0.2	0.9	0.260	1.1	2.5	0.452	0.436	0.4	0.7	0.615	0.6	1.4	2.0	1.6	0.5	0.2	0.9	1.1	3.4	2021-2022	2020	False	False	666.0	9.7	0.457	0.255	0.199	5.4	13.8	9.3	19.5	2.0	1.5	19.6	17.0	-0.4	0.7	0.3	0.020	-3.5	0.2	-3.4	-0.2	38.0	30.0	0.559	16.0	108.9	107.3	1.65	7.0	7	True	14	0.852941
3	Scottie Barnes	PF	20	TOR	Toronto Raptors (7)	60	60	35.6	6.2	12.5	0.491	0.8	2.6	0.314	5.4	10.0	0.536	0.523	2.2	3.0	0.732	2.6	5.0	7.6	3.4	1.2	0.8	1.9	2.6	15.3	2021-2022	2020	False	False	2133.0	16.4	0.552	0.203	0.238	7.6	16.1	11.6	14.3	1.6	2.3	12.1	18.9	3.0	2.3	5.3	0.119	0.5	0.3	0.8	1.5	38.0	30.0	0.559	16.0	108.9	107.3	1.65	7.0	7	True	14	0.882353
4	Khem Birch	C	29	TOR	Toronto Raptors (7)	42	17	18.3	1.8	3.7	0.474	0.0	0.2	0.000	1.8	3.5	0.500	0.474	1.0	1.4	0.746	2.2	2.2	4.5	1.0	0.6	0.5	0.5	1.9	4.6	2021-2022	2020	False	False	769.0	13.4	0.528	0.051	0.378	12.5	13.8	13.1	7.8	1.7	2.7	11.2	11.4	1.2	0.8	2.0	0.123	-1.5	0.4	-1.1	0.2	38.0	30.0	0.559	16.0	108.9	107.3	1.65	7.0	7	True	14	0.617647

	Age	G	GS	MPG	FG	FGA	FG%	3P	3PA	3P%	2P	2PA	2P%	eFG%	FT	FTA	FT%	ORB	DRB	TRB	AST	STL	BLK	TOV	PF	PTS	Decade	MPTot	PER	TS%	3PAr	FTr	ORB%	DRB%	TRB%	AST%	STL%	BLK%	TOV%	USG%	OWS	DWS	WS	WS/48	OBPM	DBPM	BPM	VORP	W	L	W/L%	GB	PS/G	PA/G	SRS	GBC	Ranking_Conf	Ranking_League	MVP_share	Game_played_prop
count	42.000000	42.000000	42.000000	42.000000	42.000000	42.000000	42.000000	42.000000	42.000000	42.000000	42.000000	42.000000	42.000000	42.000000	42.000000	42.000000	42.000000	42.000000	42.000000	42.000000	42.000000	42.000000	42.000000	42.000000	42.000000	42.000000	42.000000	42.000000	42.000000	42.000000	42.000000	42.000000	42.000000	42.000000	42.000000	42.000000	42.000000	42.000000	42.000000	42.000000	42.000000	42.000000	42.000000	42.000000	42.000000	42.000000	42.000000	42.000000	42.000000	42.000000	42.000000	42.000000	42.000000	42.000000	42.000000	42.000000	42.000000	42.000000	42.000000	42.000000
mean	27.833333	77.642857	75.547619	37.421429	9.773810	19.223810	0.510167	1.028571	2.766667	0.284429	8.757143	16.447619	0.534381	0.537429	6.311905	7.871429	0.811357	2.061905	6.740476	8.788095	6.230952	1.569048	1.154762	3.247619	2.395238	26.888095	1996.190476	2909.238095	27.783333	0.594857	0.146643	0.410952	6.352381	19.542857	13.214286	29.007143	2.104762	2.071429	12.747619	30.316667	10.776190	5.059524	15.835714	0.262048	6.961905	1.919048	8.885714	7.969048	59.976190	20.428571	0.745429	2.547619	106.847619	99.800000	6.781429	1.595238	1.500000	2.095238	0.885786	0.965493
std	2.853467	6.551192	13.588378	2.474705	1.510461	2.973198	0.038808	1.143014	2.774155	0.142665	2.058906	3.915213	0.041743	0.039912	1.519821	2.037488	0.080123	1.309990	2.327676	3.294060	2.768595	0.571471	0.952328	0.572644	0.522168	3.732403	12.287686	344.550949	2.835482	0.033258	0.143796	0.099780	3.595677	6.566922	4.696266	12.037652	0.748642	1.565214	2.801240	4.070232	2.409413	1.403143	2.715163	0.037423	1.636765	1.308553	2.394259	1.881033	7.369972	5.700419	0.069232	4.696664	7.239096	6.882817	2.285278	3.914074	1.087624	2.021947	0.099308	0.039231
min	22.000000	49.000000	0.000000	30.400000	5.700000	11.400000	0.420000	0.000000	0.000000	0.000000	4.500000	8.500000	0.441000	0.447000	2.800000	3.200000	0.524000	0.600000	2.600000	3.300000	1.300000	0.500000	0.100000	2.100000	1.400000	15.500000	1980.000000	1832.000000	22.000000	0.518000	0.001000	0.250000	1.800000	7.600000	5.200000	5.100000	0.600000	0.200000	7.700000	20.500000	6.100000	1.200000	9.600000	0.190000	3.500000	-1.800000	3.700000	4.000000	37.000000	9.000000	0.561000	0.000000	93.300000	86.800000	-0.390000	0.000000	1.000000	1.000000	0.658000	0.863014
25%	26.000000	76.250000	76.000000	36.225000	8.850000	17.225000	0.493500	0.100000	0.350000	0.240250	7.775000	14.850000	0.507000	0.516000	5.150000	6.575000	0.765500	1.150000	4.800000	6.300000	4.325000	1.150000	0.500000	3.000000	2.100000	24.525000	1990.000000	2806.500000	25.375000	0.576000	0.018750	0.329750	3.700000	14.375000	9.500000	20.150000	1.550000	0.925000	11.100000	28.100000	9.500000	4.150000	14.325000	0.232750	5.625000	1.325000	7.275000	7.100000	57.250000	17.000000	0.707000	0.000000	100.875000	92.825000	5.755000	0.000000	1.000000	1.000000	0.842250	0.951220
50%	27.500000	80.000000	79.500000	37.650000	9.900000	19.550000	0.508500	0.850000	2.400000	0.317000	8.800000	16.600000	0.530000	0.531000	6.400000	8.050000	0.823000	1.700000	6.750000	8.000000	5.750000	1.600000	0.850000	3.200000	2.500000	27.200000	2000.000000	2988.500000	27.300000	0.598000	0.128000	0.426000	5.350000	20.050000	12.850000	25.950000	2.200000	1.700000	12.600000	30.050000	10.600000	5.050000	15.750000	0.267500	6.950000	2.000000	9.000000	7.850000	62.000000	20.000000	0.756000	0.000000	107.500000	101.150000	6.700000	0.000000	1.000000	1.000000	0.913000	0.975610
75%	29.000000	81.750000	81.000000	38.875000	10.650000	21.025000	0.529500	1.400000	4.300000	0.390250	9.775000	19.075000	0.554000	0.556250	7.525000	9.875000	0.871750	2.700000	8.500000	10.800000	7.600000	1.800000	1.500000	3.475000	2.700000	29.700000	2010.000000	3137.750000	30.600000	0.617500	0.240250	0.480750	8.425000	24.125000	17.025000	38.450000	2.400000	2.700000	13.800000	32.850000	12.100000	6.075000	17.775000	0.291250	8.300000	2.650000	10.475000	9.150000	65.000000	23.750000	0.793000	3.000000	112.325000	105.375000	8.167500	0.000000	1.750000	2.000000	0.967000	1.000000
max	35.000000	82.000000	82.000000	42.000000	13.000000	25.500000	0.604000	5.100000	11.200000	0.454000	13.000000	23.700000	0.641000	0.630000	8.800000	10.500000	0.921000	6.900000	11.400000	15.300000	12.800000	3.200000	3.700000	5.400000	3.600000	35.000000	2020.000000	3398.000000	31.900000	0.669000	0.554000	0.604000	17.500000	34.800000	22.100000	57.300000	3.900000	5.900000	20.300000	41.700000	15.200000	8.000000	21.200000	0.322000	10.300000	4.200000	13.200000	12.500000	73.000000	36.000000	0.890000	20.000000	118.700000	110.100000	11.800000	20.000000	6.000000	10.000000	1.000000	1.000000

	Age	G	GS	MPG	FG	FGA	FG%	3P	3PA	3P%	2P	2PA	2P%	eFG%	FT	FTA	FT%	ORB	DRB	TRB	AST	STL	BLK	TOV	PF	PTS	Decade	MPTot	PER	TS%	3PAr	FTr	ORB%	DRB%	TRB%	AST%	STL%	BLK%	TOV%	USG%	OWS	DWS	WS	WS/48	OBPM	DBPM	BPM	VORP	W	L	W/L%	GB	PS/G	PA/G	SRS	GBC	Ranking_Conf	Ranking_League	MVP_share	Game_played_prop
count	15721.000000	15721.000000	15721.000000	15721.000000	15721.000000	15721.000000	15721.000000	15721.000000	15721.000000	15721.000000	15721.000000	15721.00000	15721.000000	15721.000000	15721.000000	15721.000000	15721.000000	15721.000000	15721.000000	15721.000000	15721.000000	15721.000000	15721.000000	15721.000000	15721.000000	15721.000000	15721.000000	15721.000000	15721.000000	15721.000000	15721.000000	15721.000000	15721.000000	15721.000000	15721.000000	15721.000000	15721.000000	15721.000000	15721.000000	15721.000000	15721.000000	15721.000000	15721.000000	15721.000000	15721.000000	15721.000000	15721.000000	15721.000000	15721.000000	15721.000000	15721.000000	15721.000000	15721.000000	15721.000000	15721.000000	15721.000000	15721.000000	15721.000000	15721.000000	15721.000000
mean	26.606068	54.716049	26.438776	20.818351	3.273284	7.153750	0.443225	0.449342	1.299491	0.216432	2.823173	5.85361	0.465212	0.472770	1.631875	2.180695	0.703776	1.067954	2.615661	3.682189	1.951269	0.688843	0.435468	1.290484	1.973284	8.626506	1997.826474	1307.216716	12.909541	0.511965	0.178847	0.313998	6.112041	14.040506	10.079683	13.162477	1.636168	1.523465	14.511819	18.940494	1.458228	1.348159	2.806984	0.074446	-1.396635	-0.164016	-1.561148	0.684079	40.007506	40.089244	0.499376	22.367534	102.400235	102.417887	-0.017059	19.574327	7.539469	14.566694	0.004474	0.682367
std	4.116044	25.395601	29.907031	10.198886	2.307366	4.756186	0.096266	0.642560	1.695272	0.182918	2.150338	4.22478	0.102893	0.098782	1.448698	1.830411	0.186494	0.881224	1.839808	2.594547	1.898318	0.484389	0.519746	0.836699	0.864625	6.158790	11.952063	927.262974	6.163956	0.095135	0.202057	0.218362	4.914280	6.554954	4.997322	9.454836	0.980640	1.752451	6.344038	5.481621	2.122491	1.251124	3.048971	0.100795	4.013899	1.809331	5.003272	1.371134	12.753868	12.712884	0.153963	12.973177	7.250777	7.312218	4.486858	13.091164	4.118828	8.236023	0.042970	0.310842
min	18.000000	1.000000	0.000000	0.000000	0.000000	0.000000	0.000000	0.000000	0.000000	0.000000	0.000000	0.00000	0.000000	0.000000	0.000000	0.000000	0.000000	0.000000	0.000000	0.000000	0.000000	0.000000	0.000000	0.000000	0.000000	0.000000	1980.000000	0.000000	-90.600000	0.000000	0.000000	0.000000	0.000000	0.000000	0.000000	0.000000	0.000000	0.000000	0.000000	0.000000	-3.300000	-1.000000	-2.100000	-2.519000	-68.900000	-31.100000	-92.100000	-2.600000	7.000000	9.000000	0.106000	0.000000	81.900000	83.400000	-14.680000	0.000000	1.000000	1.000000	0.000000	0.012195
25%	23.000000	36.000000	0.000000	12.300000	1.400000	3.400000	0.407000	0.000000	0.000000	0.000000	1.100000	2.60000	0.432000	0.442000	0.600000	0.900000	0.653000	0.400000	1.300000	1.800000	0.600000	0.300000	0.100000	0.700000	1.300000	3.800000	1990.000000	444.000000	10.100000	0.482000	0.007000	0.197000	2.500000	9.200000	6.100000	6.500000	1.100000	0.400000	11.100000	15.400000	0.000000	0.300000	0.300000	0.040000	-3.100000	-1.000000	-3.300000	-0.100000	30.000000	30.000000	0.378000	12.000000	96.900000	97.000000	-3.160000	9.000000	4.000000	7.000000	0.000000	0.463415
50%	26.000000	63.000000	11.000000	20.500000	2.700000	6.000000	0.448000	0.100000	0.500000	0.259000	2.200000	4.70000	0.474000	0.482000	1.200000	1.600000	0.748000	0.800000	2.200000	3.000000	1.300000	0.600000	0.300000	1.100000	2.000000	7.100000	2000.000000	1254.000000	13.000000	0.523000	0.086000	0.284000	5.200000	13.100000	9.300000	10.400000	1.500000	1.000000	13.700000	18.600000	0.700000	1.000000	1.900000	0.084000	-1.200000	-0.200000	-1.300000	0.100000	41.000000	40.000000	0.512000	21.000000	101.900000	102.400000	0.160000	18.000000	7.000000	14.000000	0.000000	0.792683
75%	29.000000	77.000000	54.000000	29.500000	4.700000	10.200000	0.490000	0.700000	2.100000	0.356000	4.000000	8.30000	0.511000	0.519000	2.200000	3.000000	0.812000	1.500000	3.500000	4.900000	2.600000	0.900000	0.600000	1.800000	2.600000	12.300000	2010.000000	2067.000000	16.000000	0.558000	0.321000	0.389000	8.900000	18.200000	13.500000	17.700000	2.100000	2.100000	17.000000	22.200000	2.400000	2.000000	4.400000	0.121000	0.600000	0.700000	0.700000	1.100000	50.000000	49.000000	0.610000	32.000000	107.700000	107.500000	3.230000	29.000000	11.000000	21.000000	0.000000	0.951220
max	44.000000	82.000000	82.000000	43.700000	13.400000	27.800000	1.000000	5.300000	13.200000	1.000000	13.200000	27.00000	1.000000	1.500000	10.300000	13.100000	1.000000	7.000000	12.300000	18.700000	14.500000	3.700000	5.600000	5.700000	6.000000	37.100000	2020.000000	3533.000000	133.800000	1.500000	1.000000	6.000000	100.000000	100.000000	86.400000	100.000000	24.200000	77.800000	100.000000	100.000000	14.700000	9.100000	20.000000	2.712000	199.400000	60.700000	242.200000	11.400000	73.000000	72.000000	0.890000	63.000000	126.500000	130.800000	11.800000	56.000000	15.000000	30.000000	0.832000	1.000000

	ORB	DRB	AST	STL	BLK	3P	OWS	DWS	PTS	MVP	counts
Cluster
0	1.672283	4.280565	3.399674	1.101804	0.728717	0.679109	3.850261	2.770957	15.263391	0.00913	4600
1	1.047965	2.554253	1.900262	0.696472	0.410533	0.501362	0.848803	0.988699	8.325939	0.00000	5725
2	0.585473	1.303825	0.812817	0.338286	0.219217	0.204689	0.148363	0.551710	3.469842	0.00000	5438

	Predicted MVP	Actual MVP	Label
year
1980	Larry Bird	Kareem Abdul-Jabbar	incorrect
1981	Julius Erving	Julius Erving	correct
1982	Magic Johnson	Moses Malone	incorrect
1983	Larry Bird	Moses Malone	incorrect
1984	Larry Bird	Larry Bird	correct
1985	Larry Bird	Larry Bird	correct
1986	Larry Bird	Larry Bird	correct
1987	Michael Jordan	Magic Johnson	incorrect
1988	Michael Jordan	Michael Jordan	correct
1989	Michael Jordan	Magic Johnson	incorrect
1990	Michael Jordan	Magic Johnson	incorrect
1991	Michael Jordan	Michael Jordan	correct
1992	Michael Jordan	Michael Jordan	correct
1993	Michael Jordan	Charles Barkley	incorrect
1994	Hakeem Olajuwon	Hakeem Olajuwon	correct
1995	David Robinson	David Robinson	correct
1996	Michael Jordan	Michael Jordan	correct
1997	Michael Jordan	Karl Malone	incorrect
1998	Michael Jordan	Michael Jordan	correct
1999	Shaquille O'Neal	Karl Malone	incorrect
2000	Shaquille O'Neal	Shaquille O'Neal	correct
2001	Shaquille O'Neal	Allen Iverson	incorrect
2002	Tim Duncan	Tim Duncan	correct
2003	Tim Duncan	Tim Duncan	correct
2004	Tim Duncan	Kevin Garnett	incorrect
2005	Kevin Garnett	Steve Nash	incorrect
2006	LeBron James	Steve Nash	incorrect
2007	Kobe Bryant	Dirk Nowitzki	incorrect
2008	Chris Paul	Kobe Bryant	incorrect
2009	LeBron James	LeBron James	correct
2010	LeBron James	LeBron James	correct
2011	LeBron James	Derrick Rose	incorrect
2012	LeBron James	LeBron James	correct
2013	Kevin Durant	LeBron James	incorrect
2014	Kevin Durant	Kevin Durant	correct
2015	James Harden	Stephen Curry	incorrect
2016	LeBron James	Stephen Curry	incorrect
2017	Russell Westbrook	Russell Westbrook	correct
2018	James Harden	James Harden	correct
2019	James Harden	Giannis Antetokounmpo	incorrect
2020	James Harden	Giannis Antetokounmpo	incorrect
2021	Nikola Jokić	Nikola Jokić	correct

MVP Player Prediction¶

Table of Contents¶

Import data/modules ¶

Exploratory Data Analysis ¶

Features Engineering/Preprocessing ¶

Clustering¶

Modeling ¶

Predicting MVP eligibility ¶

SVM¶

Predicting MVP Candidates ¶

KNN¶

XGBooster¶

Get Prediction Accuracy by LOOCV¶

Leave One Out Cross Validation¶

Predicting MVP in the current season 2021-2022 ¶

Random Forest Prediction¶

KNN¶

XGBoost¶

	Player	Pos	Age	Tm	Team	G	GS	MPG	FG	FGA	FG%	3P	3PA	3P%	2P	2PA	2P%	eFG%	FT	FTA	FT%	ORB	DRB	TRB	AST	STL	BLK	TOV	PF	PTS	Season	Decade	Trade	MPTot	PER	TS%	3PAr	FTr	ORB%	DRB%	TRB%	AST%	STL%	BLK%	TOV%	USG%	OWS	DWS	WS	WS/48	OBPM	DBPM	BPM	VORP	MVP	PastMVP	W	L	W/L%	GB	PS/G	PA/G	SRS	Playoffs	Ranking_Conf	Ranking_League	MVP_share	Game_played_prop	ShareYN
0	Kareem Abdul-Jabbar	C	32	LAL	Los Angeles Lakers	82	NaN	38.3	10.2	16.9	0.604	0.0	0.0	0.00	10.2	16.9	0.604	0.604	4.4	5.8	0.765	2.3	8.5	10.8	4.5	1.0	3.4	3.6	2.6	24.8	1979-1980	1980	False	3143.0	25.3	0.639	0.001	0.344	7.2	22.2	15.4	16.5	1.2	4.6	15.7	24.1	9.5	5.3	14.8	0.227	4.8	2.4	7.2	7.3	True	True	60	22	0.732	1.0	115.1	109.2	5.4	True	1.0	2.0	0.665	1.000000	True
1	Marty Byrnes	SF	23	LAL	Los Angeles Lakers	32	NaN	6.1	0.8	1.6	0.500	0.0	0.0	NaN	0.8	1.6	0.500	0.500	0.4	0.5	0.867	0.3	0.6	0.8	0.4	0.2	0.0	0.7	1.0	2.0	1979-1980	1980	False	194.0	6.3	0.557	0.000	0.300	5.5	9.3	7.6	7.9	1.2	0.3	28.0	16.2	-0.1	0.2	0.0	0.009	-4.4	-0.9	-5.2	-0.2	False	False	60	22	0.732	1.0	115.1	109.2	5.4	True	1.0	2.0	0.000	0.390244	False
2	Jim Chones	C	30	LAL	Los Angeles Lakers	82	NaN	29.2	4.5	9.3	0.489	0.0	0.0	0.00	4.5	9.2	0.491	0.489	1.5	2.1	0.740	1.7	5.1	6.9	1.8	0.7	0.8	2.1	3.3	10.6	1979-1980	1980	False	2394.0	11.3	0.521	0.003	0.222	7.1	17.6	12.8	7.6	1.1	1.4	17.3	16.9	0.7	2.8	3.5	0.071	-2.0	-0.3	-2.4	-0.2	False	False	60	22	0.732	1.0	115.1	109.2	5.4	True	1.0	2.0	0.000	1.000000	False
3	Michael Cooper	SG	23	LAL	Los Angeles Lakers	82	NaN	24.1	3.7	7.0	0.524	0.1	0.2	0.25	3.6	6.8	0.534	0.529	1.4	1.7	0.776	1.2	1.6	2.8	2.7	1.0	0.5	1.7	2.6	8.8	1979-1980	1980	False	1973.0	12.5	0.563	0.035	0.247	6.1	6.5	6.3	13.5	2.0	1.0	18.1	15.9	2.2	1.8	4.0	0.097	-0.4	0.3	-0.1	0.9	False	False	60	22	0.732	1.0	115.1	109.2	5.4	True	1.0	2.0	0.000	1.000000	False
4	Spencer Haywood	PF	30	LAL	Los Angeles Lakers	76	NaN	20.3	3.8	7.8	0.487	0.0	0.1	0.25	3.8	7.7	0.489	0.488	2.1	2.7	0.772	1.7	2.8	4.6	1.2	0.5	0.8	1.8	2.6	9.7	1979-1980	1980	False	1544.0	13.9	0.540	0.007	0.349	10.2	13.9	12.2	7.6	1.0	1.9	16.4	21.2	1.4	1.7	3.1	0.095	-1.3	-0.9	-2.3	-0.1	False	False	60	22	0.732	1.0	115.1	109.2	5.4	True	1.0	2.0	0.000	0.926829	False

	Player	Prediction	confidence_score
0	Giannis Antetokounmpo	1.0	12.064325
1	Nikola Jokić	1.0	11.829144
2	Joel Embiid	1.0	10.229131
3	Luka Dončić	1.0	8.524647
4	Ja Morant	1.0	7.366794
5	LeBron James	1.0	6.922947
6	Stephen Curry	1.0	6.875250
7	Chris Paul	1.0	6.748942
8	DeMar DeRozan	1.0	6.588772
9	Jimmy Butler	1.0	6.410305
10	Trae Young	1.0	6.330801
11	Karl-Anthony Towns	1.0	6.252892
12	Kevin Durant	1.0	6.194515
13	Jayson Tatum	1.0	5.872098
14	Devin Booker	1.0	5.516034

	Player	Prediction
309	Nikola Jokić	13.085095
75	Giannis Antetokounmpo	12.817300
317	Joel Embiid	8.841221
484	Luka Dončić	4.602219
34	Ja Morant	3.891855

	Player	Prediction
309	Nikola Jokić	9.346072
75	Giannis Antetokounmpo	8.929789
317	Joel Embiid	4.473491
222	Rudy Gobert	3.837105
396	Karl-Anthony Towns	2.425572

	Player	Prediction
75	Giannis Antetokounmpo	12.060459
309	Nikola Jokić	11.398022
317	Joel Embiid	7.923958
34	Ja Morant	5.272460
484	Luka Dončić	4.146081

	Player	Prediction
309	Nikola Jokić	9.928906
75	Giannis Antetokounmpo	8.761899
317	Joel Embiid	6.278138
484	Luka Dončić	6.038182
435	Stephen Curry	5.061064

MVP Player Prediction¶

Table of Contents¶

Import data/modules ¶

Exploratory Data Analysis ¶

Features Engineering/Preprocessing ¶

Clustering¶

Modeling ¶

Predicting MVP eligibility ¶

SVM¶

Predicting MVP Candidates ¶

Forecasting 2022 MVP (Predicting MVP Share) ¶

Models for predicting MVP share ¶

KNN¶

XGBooster¶

Get Prediction Accuracy by LOOCV¶

Leave One Out Cross Validation¶

Predicting MVP in the current season 2021-2022 ¶

Random Forest Prediction¶

KNN¶

XGBoost¶